Deterministic chart annotation

A model can find the story.
It cannot place the arrow.

This engine reads 927 metro-level housing series and asks the model for exactly one thing: a series and a date. Every pixel below is computed. The label finds the space where it cannot cross a trendline.

The bottleneck is not the writing.

Language models find the anomaly and draft the copy in an afternoon. What breaks at volume is the picture: a label through a trendline, an arrow off the canvas, a figure the model invented. That layer has to be arithmetic, and it has to be someone's job.

15 minutes to talk it through

Frederic de Lavenne de Choulot ยท The AI Pipe